Source: Azeem Azhar, Exponential View
The spike in B200 GPU costs signals a hard constraint on AI scaling: physical chip supply cannot keep pace with enterprise demand, pushing compute access into a landlord-tenant dynamic where infrastructure providers capture margin instead of chip makers. Companies are willing to pay exponentially more for immediate access to training infrastructure, a real-time pricing signal that deployment timelines are accelerating faster than supply chains can respond. Whoever controls GPU allocation in the next 18 months owns a significant choke point in the AI stack.